Each reduce task take the role of DFS client by using specified output folder. The placement of reduce output result judged by NameNode block placement policy. One thing I can confirm, as the cluster inner slave, the first block reduce task produced in one DataNode can be placed into local filesystem.
On Thu, Jun 16, 2011 at 3:21 AM, Pedro Costa <psdc1...@gmail.com> wrote: > Hi, > > I'm running an MR application that produces an output that is saved in > HDFS. My application has 5 slave nodes (so it has also 5 data nodes). > The hdfs file replication factor is 1. I want from my application, or > from the Hadoop MR source code tell which data node my result should > be. For example, if I've datanode1-5, I would like to say to the name > node to save the reduce output result data in a specific datanode. > > Is it possible? > > > Thanks, >
